Model Card for gemma-3n-FoodExtract-Vision-v1

This model is a fine-tuned version of google/gemma-3-4b-it specialized in structured food extraction. It analyzes images to determine if they contain food, generates a short title, and extracts lists of visible food and drink items in a specific JSON format.

Training Strategy: This model was trained using PEFT (LoRA) on the Gemma 3 architecture. Gemma 3 is a native multimodal model, meaning it processes image and text tokens in the same transformer block.

Quick start

This model relies on a specific system prompt and user prompt structure to output the correct JSON format.

import torch
from transformers import AutoProcessor, AutoModelForImageTextToText
from peft import PeftModel
from PIL import Image
import requests

# 1) Load base model + processor, then apply the LoRA adapter
base_model_id = "google/gemma-3n-E2B-it"
adapter_model_id = "berkeruveyik/gemma-3n-FoodExtract-Vision-v1"

processor = AutoProcessor.from_pretrained(base_model_id)
model = AutoModelForImageTextToText.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model = PeftModel.from_pretrained(model, adapter_model_id)
model.eval()

# 2) Prompts
SYSTEM_MESSAGE = """You are an expert food and drink image extractor.
You provide structured data to visual inputs classifying them as edible food/drink or not.
as well as titling the image with a simple simple food/drink related caption.
Finally you extract any and all visible food/drink items to lists."""

USER_PROMPT = """Classify the given input image into food or not, and if edible food or drink items are present, extract them into lists. If no food/drink items are visible, return an empty list.

Only return valid JSON in the following form:

```json
{
  "is_food": 0,
  "image_title": "",
  "food_items": [],
  "drink_items": []
}
```"""

# 3) Load image (use a direct URL, not Markdown)
image_url = "https://img.freepik.com/free-psd/roasted-chicken-dinner-platter-delicious-feast_632498-25445.jpg?semt=ais_hybrid&w=740&q=80"
resp = requests.get(image_url, stream=True, headers={"User-Agent": "Mozilla/5.0"})
resp.raise_for_status()
image = Image.open(resp.raw).convert("RGB")

# 4) Prepare inputs:
#    - Create chat-formatted text with apply_chat_template (tokenize=False)
#    - Then run the processor on (text + image) to get tensors
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image},
            {"type": "text", "text": SYSTEM_MESSAGE + "\n\n" + USER_PROMPT},
        ],
    }
]

text = processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=False,
)

inputs = processor(
    text=text,
    images=image,
    return_tensors="pt",
)

# Move tensors to the same device as the model
inputs = {k: v.to(model.device) for k, v in inputs.items()}

# Cast float32 tensors to model dtype (bfloat16) when needed
inputs = {
    k: (v.to(dtype=model.dtype) if torch.is_floating_point(v) and v.dtype == torch.float32 else v)
    for k, v in inputs.items()
}

# 5) Generate
with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=256)

# 6) Decode only the newly generated tokens
prompt_len = inputs["input_ids"].shape[1]
output_text = processor.batch_decode(
    generated_ids[:, prompt_len:],
    skip_special_tokens=True
)[0]

print(output_text)

Training procedure

This model was trained using Low-Rank Adaptation (LoRA) on the Gemma 3 architecture.

Training Hyperparameters

The following hyperparameters were used during training:

• Training Regime: PEFT (LoRA)
• Num train epochs: 3
• Learning rate: 2e-4
• Batch size per device: 4
• Gradient accumulation steps: 4
• Optimizer: adamw_torch
• LoRA Rank (r): 64
• LoRA Alpha: 128
• LoRA Dropout: 0.05
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• BF16: True

Framework versions

• TRL: 0.27.2
• Transformers: 4.49.0
• Pytorch: 2.9.0+cu126
• Datasets: 4.0.0
• Tokenizers: 0.22.2
• PEFT: 0.10.0

Citations

Cite Gemma 3 and PEFT as:

@article{gemma32024,
    title={Gemma 3: Multimodal Native Open Models},
    url={[https://storage.googleapis.com/deepmind-media/gemma/gemma-3-report.pdf](https://storage.googleapis.com/deepmind-media/gemma/gemma-3-report.pdf)},
    DOI={10.48550/arXiv.2408.00118},
    publisher={Google},
    author={Gemma Team, Google DeepMind},
    year={2024}
}

@misc{mangrulkar2022peft,
    title        = {PEFT: State-of-the-art Parameter-Efficient Fine-Tuning methods},
    author       = {Sourab Mangrulkar and Sylvain Gugger and Lysandre Debut and Younes Belkada and Sayak Paul and Benjamin Bossan},
    year         = {2022},
    publisher    = {Hugging Face},
    howpublished = {\url{[https://github.com/huggingface/peft](https://github.com/huggingface/peft)}},
}